Robust reconstruction of textured surfaces from 3D Point Clouds

Abstract

There are three main approaches for 3D building reconstruction from images. Laser Scanning is accurate, but expensive and limited by the laser?s range. SfM recovers 3D data and camera parameters from a sequence of images. MVS methods, especially patch-based MVS (PMVS), can achieve higher accuracy and recover denser 3D point cloud than SfM. But, the recovered point cloud can be sparse in regions that lack features for matching, making recovery of complete surfaces difficult. This thesis presents a robust reconstruction of textured surfaces from 3D point clouds given by patch-based MVS method. To a good first approximation, a building?s surfaces can be modeled by either planes or curve surfaces which are fitted into the point cloud. 3D points are resampled on the fitted surfaces in an orderly pattern, whose colors are retrieved from input images. This approach is thus simple, inexpensive and effective for recovering textured surfaces of buildings.